2020 Theses Doctoral
Extracting cosmological information from small scales in weak gravitational lensing data
This work is concerned with how to extract information encoded in small scales of non-Gaussian fields, with the purpose of learning about cosmology using weak gravitational lensing. We do so by comparing different methods on simulated data sets. The topic is relevant, for upcoming galaxy surveys will map the late evolution of the matter density field, which is non-Gaussian, with an unprecedented level of detail, and any improvement on the analysis techniques will increase the experiments' scientific return.
First, we investigate some non-Gaussian observables used in the weak lensing community. We analyze to what extent they are sensitive to the background expansion of the universe, and to what extent to the evolution of the structures responsible for the lensing. We then focus our attention on one such statistic, lensing peaks, and assess the performance of a simple halo-based model that has been proposed to forecast their abundance. We find some shortcomings of that semi-analytic approach, and proceed to review some minimal requirements for numerical simulations used to forecast non-Gaussian statistics, to reduce their computational cost while fulfilling the accuracy and precision required by future experiments.
Second, we propose a novel measurement, that of the temperature dipole induced on the cosmic microwave background induced by the rotation of ionized gas around galaxies, as an additional observation to help constrain the distribution of baryonic matter on the smallest scales probed by WL experiments. The uncertainty in this distribution is a major theoretical systematic for future surveys.
Third, we show how deep neural networks can be used to map pixel-level data into the cosmological parameters of interest, by-passing the previous compression step of measuring pre-designed statistics. We provide the first (simulation-based) credible contours based on neural networks applied to weak lensing data, and discuss how to interpret these models.
- ZorrillaMatilla_columbia_0054D_16081.pdf application/pdf 7.25 MB Download File
More About This Work
- Academic Units
- Thesis Advisors
- Haiman, Zoltan
- Ph.D., Columbia University
- Published Here
- July 30, 2020